2 * xt_hashlimit - Netfilter module to limit the number of packets per time
3 * separately for each hashbucket (sourceip/sourceport/dstip/dstport)
5 * (C) 2003-2004 by Harald Welte <laforge@netfilter.org>
6 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
7 * Copyright © CC Computer Consultants GmbH, 2007 - 2008
9 * Development of this code was funded by Astaro AG, http://www.astaro.com/
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/module.h>
13 #include <linux/spinlock.h>
14 #include <linux/random.h>
15 #include <linux/jhash.h>
16 #include <linux/slab.h>
17 #include <linux/vmalloc.h>
18 #include <linux/proc_fs.h>
19 #include <linux/seq_file.h>
20 #include <linux/list.h>
21 #include <linux/skbuff.h>
25 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
26 #include <linux/ipv6.h>
30 #include <net/net_namespace.h>
31 #include <net/netns/generic.h>
33 #include <linux/netfilter/x_tables.h>
34 #include <linux/netfilter_ipv4/ip_tables.h>
35 #include <linux/netfilter_ipv6/ip6_tables.h>
36 #include <linux/netfilter/xt_hashlimit.h>
37 #include <linux/mutex.h>
38 #include <linux/kernel.h>
40 MODULE_LICENSE("GPL");
41 MODULE_AUTHOR("Harald Welte <laforge@netfilter.org>");
42 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
43 MODULE_DESCRIPTION("Xtables: per hash-bucket rate-limit match");
44 MODULE_ALIAS("ipt_hashlimit");
45 MODULE_ALIAS("ip6t_hashlimit");
47 struct hashlimit_net
{
48 struct hlist_head htables
;
49 struct proc_dir_entry
*ipt_hashlimit
;
50 struct proc_dir_entry
*ip6t_hashlimit
;
53 static unsigned int hashlimit_net_id
;
54 static inline struct hashlimit_net
*hashlimit_pernet(struct net
*net
)
56 return net_generic(net
, hashlimit_net_id
);
59 /* need to declare this at the top */
60 static const struct file_operations dl_file_ops_v2
;
61 static const struct file_operations dl_file_ops_v1
;
62 static const struct file_operations dl_file_ops
;
71 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
83 /* static / read-only parts in the beginning */
84 struct hlist_node node
;
85 struct dsthash_dst dst
;
87 /* modified structure members in the end */
89 unsigned long expires
; /* precalculated expiry time */
91 unsigned long prev
; /* last modification */
99 u_int32_t interval
, prev_window
;
100 u_int64_t current_rate
;
109 struct xt_hashlimit_htable
{
110 struct hlist_node node
; /* global list of all htables */
113 bool rnd_initialized
;
115 struct hashlimit_cfg3 cfg
; /* config */
117 /* used internally */
118 spinlock_t lock
; /* lock for list_head */
119 u_int32_t rnd
; /* random seed for hash */
120 unsigned int count
; /* number entries in table */
121 struct delayed_work gc_work
;
124 struct proc_dir_entry
*pde
;
128 struct hlist_head hash
[0]; /* hashtable itself */
132 cfg_copy(struct hashlimit_cfg3
*to
, const void *from
, int revision
)
135 struct hashlimit_cfg1
*cfg
= (struct hashlimit_cfg1
*)from
;
137 to
->mode
= cfg
->mode
;
139 to
->burst
= cfg
->burst
;
140 to
->size
= cfg
->size
;
142 to
->gc_interval
= cfg
->gc_interval
;
143 to
->expire
= cfg
->expire
;
144 to
->srcmask
= cfg
->srcmask
;
145 to
->dstmask
= cfg
->dstmask
;
146 } else if (revision
== 2) {
147 struct hashlimit_cfg2
*cfg
= (struct hashlimit_cfg2
*)from
;
149 to
->mode
= cfg
->mode
;
151 to
->burst
= cfg
->burst
;
152 to
->size
= cfg
->size
;
154 to
->gc_interval
= cfg
->gc_interval
;
155 to
->expire
= cfg
->expire
;
156 to
->srcmask
= cfg
->srcmask
;
157 to
->dstmask
= cfg
->dstmask
;
158 } else if (revision
== 3) {
159 memcpy(to
, from
, sizeof(struct hashlimit_cfg3
));
167 static DEFINE_MUTEX(hashlimit_mutex
); /* protects htables list */
168 static struct kmem_cache
*hashlimit_cachep __read_mostly
;
170 static inline bool dst_cmp(const struct dsthash_ent
*ent
,
171 const struct dsthash_dst
*b
)
173 return !memcmp(&ent
->dst
, b
, sizeof(ent
->dst
));
177 hash_dst(const struct xt_hashlimit_htable
*ht
, const struct dsthash_dst
*dst
)
179 u_int32_t hash
= jhash2((const u32
*)dst
,
180 sizeof(*dst
)/sizeof(u32
),
183 * Instead of returning hash % ht->cfg.size (implying a divide)
184 * we return the high 32 bits of the (hash * ht->cfg.size) that will
185 * give results between [0 and cfg.size-1] and same hash distribution,
186 * but using a multiply, less expensive than a divide
188 return reciprocal_scale(hash
, ht
->cfg
.size
);
191 static struct dsthash_ent
*
192 dsthash_find(const struct xt_hashlimit_htable
*ht
,
193 const struct dsthash_dst
*dst
)
195 struct dsthash_ent
*ent
;
196 u_int32_t hash
= hash_dst(ht
, dst
);
198 if (!hlist_empty(&ht
->hash
[hash
])) {
199 hlist_for_each_entry_rcu(ent
, &ht
->hash
[hash
], node
)
200 if (dst_cmp(ent
, dst
)) {
201 spin_lock(&ent
->lock
);
208 /* allocate dsthash_ent, initialize dst, put in htable and lock it */
209 static struct dsthash_ent
*
210 dsthash_alloc_init(struct xt_hashlimit_htable
*ht
,
211 const struct dsthash_dst
*dst
, bool *race
)
213 struct dsthash_ent
*ent
;
215 spin_lock(&ht
->lock
);
217 /* Two or more packets may race to create the same entry in the
218 * hashtable, double check if this packet lost race.
220 ent
= dsthash_find(ht
, dst
);
222 spin_unlock(&ht
->lock
);
227 /* initialize hash with random val at the time we allocate
228 * the first hashtable entry */
229 if (unlikely(!ht
->rnd_initialized
)) {
230 get_random_bytes(&ht
->rnd
, sizeof(ht
->rnd
));
231 ht
->rnd_initialized
= true;
234 if (ht
->cfg
.max
&& ht
->count
>= ht
->cfg
.max
) {
235 /* FIXME: do something. question is what.. */
236 net_err_ratelimited("max count of %u reached\n", ht
->cfg
.max
);
239 ent
= kmem_cache_alloc(hashlimit_cachep
, GFP_ATOMIC
);
241 memcpy(&ent
->dst
, dst
, sizeof(ent
->dst
));
242 spin_lock_init(&ent
->lock
);
244 spin_lock(&ent
->lock
);
245 hlist_add_head_rcu(&ent
->node
, &ht
->hash
[hash_dst(ht
, dst
)]);
248 spin_unlock(&ht
->lock
);
252 static void dsthash_free_rcu(struct rcu_head
*head
)
254 struct dsthash_ent
*ent
= container_of(head
, struct dsthash_ent
, rcu
);
256 kmem_cache_free(hashlimit_cachep
, ent
);
260 dsthash_free(struct xt_hashlimit_htable
*ht
, struct dsthash_ent
*ent
)
262 hlist_del_rcu(&ent
->node
);
263 call_rcu_bh(&ent
->rcu
, dsthash_free_rcu
);
266 static void htable_gc(struct work_struct
*work
);
268 static int htable_create(struct net
*net
, struct hashlimit_cfg3
*cfg
,
269 const char *name
, u_int8_t family
,
270 struct xt_hashlimit_htable
**out_hinfo
,
273 struct hashlimit_net
*hashlimit_net
= hashlimit_pernet(net
);
274 struct xt_hashlimit_htable
*hinfo
;
275 const struct file_operations
*fops
;
276 unsigned int size
, i
;
282 size
= (totalram_pages
<< PAGE_SHIFT
) / 16384 /
283 sizeof(struct hlist_head
);
284 if (totalram_pages
> 1024 * 1024 * 1024 / PAGE_SIZE
)
289 /* FIXME: don't use vmalloc() here or anywhere else -HW */
290 hinfo
= vmalloc(sizeof(struct xt_hashlimit_htable
) +
291 sizeof(struct hlist_head
) * size
);
296 /* copy match config into hashtable config */
297 ret
= cfg_copy(&hinfo
->cfg
, (void *)cfg
, 3);
302 hinfo
->cfg
.size
= size
;
303 if (hinfo
->cfg
.max
== 0)
304 hinfo
->cfg
.max
= 8 * hinfo
->cfg
.size
;
305 else if (hinfo
->cfg
.max
< hinfo
->cfg
.size
)
306 hinfo
->cfg
.max
= hinfo
->cfg
.size
;
308 for (i
= 0; i
< hinfo
->cfg
.size
; i
++)
309 INIT_HLIST_HEAD(&hinfo
->hash
[i
]);
313 hinfo
->family
= family
;
314 hinfo
->rnd_initialized
= false;
315 hinfo
->name
= kstrdup(name
, GFP_KERNEL
);
320 spin_lock_init(&hinfo
->lock
);
324 fops
= &dl_file_ops_v1
;
327 fops
= &dl_file_ops_v2
;
333 hinfo
->pde
= proc_create_data(name
, 0,
334 (family
== NFPROTO_IPV4
) ?
335 hashlimit_net
->ipt_hashlimit
: hashlimit_net
->ip6t_hashlimit
,
337 if (hinfo
->pde
== NULL
) {
344 INIT_DEFERRABLE_WORK(&hinfo
->gc_work
, htable_gc
);
345 queue_delayed_work(system_power_efficient_wq
, &hinfo
->gc_work
,
346 msecs_to_jiffies(hinfo
->cfg
.gc_interval
));
348 hlist_add_head(&hinfo
->node
, &hashlimit_net
->htables
);
353 static bool select_all(const struct xt_hashlimit_htable
*ht
,
354 const struct dsthash_ent
*he
)
359 static bool select_gc(const struct xt_hashlimit_htable
*ht
,
360 const struct dsthash_ent
*he
)
362 return time_after_eq(jiffies
, he
->expires
);
365 static void htable_selective_cleanup(struct xt_hashlimit_htable
*ht
,
366 bool (*select
)(const struct xt_hashlimit_htable
*ht
,
367 const struct dsthash_ent
*he
))
371 for (i
= 0; i
< ht
->cfg
.size
; i
++) {
372 struct dsthash_ent
*dh
;
373 struct hlist_node
*n
;
375 spin_lock_bh(&ht
->lock
);
376 hlist_for_each_entry_safe(dh
, n
, &ht
->hash
[i
], node
) {
377 if ((*select
)(ht
, dh
))
378 dsthash_free(ht
, dh
);
380 spin_unlock_bh(&ht
->lock
);
385 static void htable_gc(struct work_struct
*work
)
387 struct xt_hashlimit_htable
*ht
;
389 ht
= container_of(work
, struct xt_hashlimit_htable
, gc_work
.work
);
391 htable_selective_cleanup(ht
, select_gc
);
393 queue_delayed_work(system_power_efficient_wq
,
394 &ht
->gc_work
, msecs_to_jiffies(ht
->cfg
.gc_interval
));
397 static void htable_remove_proc_entry(struct xt_hashlimit_htable
*hinfo
)
399 struct hashlimit_net
*hashlimit_net
= hashlimit_pernet(hinfo
->net
);
400 struct proc_dir_entry
*parent
;
402 if (hinfo
->family
== NFPROTO_IPV4
)
403 parent
= hashlimit_net
->ipt_hashlimit
;
405 parent
= hashlimit_net
->ip6t_hashlimit
;
408 remove_proc_entry(hinfo
->name
, parent
);
411 static void htable_destroy(struct xt_hashlimit_htable
*hinfo
)
413 cancel_delayed_work_sync(&hinfo
->gc_work
);
414 htable_remove_proc_entry(hinfo
);
415 htable_selective_cleanup(hinfo
, select_all
);
420 static struct xt_hashlimit_htable
*htable_find_get(struct net
*net
,
424 struct hashlimit_net
*hashlimit_net
= hashlimit_pernet(net
);
425 struct xt_hashlimit_htable
*hinfo
;
427 hlist_for_each_entry(hinfo
, &hashlimit_net
->htables
, node
) {
428 if (!strcmp(name
, hinfo
->name
) &&
429 hinfo
->family
== family
) {
437 static void htable_put(struct xt_hashlimit_htable
*hinfo
)
439 mutex_lock(&hashlimit_mutex
);
440 if (--hinfo
->use
== 0) {
441 hlist_del(&hinfo
->node
);
442 htable_destroy(hinfo
);
444 mutex_unlock(&hashlimit_mutex
);
447 /* The algorithm used is the Simple Token Bucket Filter (TBF)
448 * see net/sched/sch_tbf.c in the linux source tree
451 /* Rusty: This is my (non-mathematically-inclined) understanding of
452 this algorithm. The `average rate' in jiffies becomes your initial
453 amount of credit `credit' and the most credit you can ever have
454 `credit_cap'. The `peak rate' becomes the cost of passing the
457 `prev' tracks the last packet hit: you gain one credit per jiffy.
458 If you get credit balance more than this, the extra credit is
459 discarded. Every time the match passes, you lose `cost' credits;
460 if you don't have that many, the test fails.
462 See Alexey's formal explanation in net/sched/sch_tbf.c.
464 To get the maximum range, we multiply by this factor (ie. you get N
465 credits per jiffy). We want to allow a rate as low as 1 per day
466 (slowest userspace tool allows), which means
467 CREDITS_PER_JIFFY*HZ*60*60*24 < 2^32 ie.
469 #define MAX_CPJ_v1 (0xFFFFFFFF / (HZ*60*60*24))
470 #define MAX_CPJ (0xFFFFFFFFFFFFFFFFULL / (HZ*60*60*24))
472 /* Repeated shift and or gives us all 1s, final shift and add 1 gives
473 * us the power of 2 below the theoretical max, so GCC simply does a
475 #define _POW2_BELOW2(x) ((x)|((x)>>1))
476 #define _POW2_BELOW4(x) (_POW2_BELOW2(x)|_POW2_BELOW2((x)>>2))
477 #define _POW2_BELOW8(x) (_POW2_BELOW4(x)|_POW2_BELOW4((x)>>4))
478 #define _POW2_BELOW16(x) (_POW2_BELOW8(x)|_POW2_BELOW8((x)>>8))
479 #define _POW2_BELOW32(x) (_POW2_BELOW16(x)|_POW2_BELOW16((x)>>16))
480 #define _POW2_BELOW64(x) (_POW2_BELOW32(x)|_POW2_BELOW32((x)>>32))
481 #define POW2_BELOW32(x) ((_POW2_BELOW32(x)>>1) + 1)
482 #define POW2_BELOW64(x) ((_POW2_BELOW64(x)>>1) + 1)
484 #define CREDITS_PER_JIFFY POW2_BELOW64(MAX_CPJ)
485 #define CREDITS_PER_JIFFY_v1 POW2_BELOW32(MAX_CPJ_v1)
487 /* in byte mode, the lowest possible rate is one packet/second.
488 * credit_cap is used as a counter that tells us how many times we can
489 * refill the "credits available" counter when it becomes empty.
491 #define MAX_CPJ_BYTES (0xFFFFFFFF / HZ)
492 #define CREDITS_PER_JIFFY_BYTES POW2_BELOW32(MAX_CPJ_BYTES)
494 static u32
xt_hashlimit_len_to_chunks(u32 len
)
496 return (len
>> XT_HASHLIMIT_BYTE_SHIFT
) + 1;
499 /* Precision saver. */
500 static u64
user2credits(u64 user
, int revision
)
502 u64 scale
= (revision
== 1) ?
503 XT_HASHLIMIT_SCALE
: XT_HASHLIMIT_SCALE_v2
;
504 u64 cpj
= (revision
== 1) ?
505 CREDITS_PER_JIFFY_v1
: CREDITS_PER_JIFFY
;
507 /* Avoid overflow: divide the constant operands first */
508 if (scale
>= HZ
* cpj
)
509 return div64_u64(user
, div64_u64(scale
, HZ
* cpj
));
511 return user
* div64_u64(HZ
* cpj
, scale
);
514 static u32
user2credits_byte(u32 user
)
517 us
*= HZ
* CREDITS_PER_JIFFY_BYTES
;
518 return (u32
) (us
>> 32);
521 static u64
user2rate(u64 user
)
524 return div64_u64(XT_HASHLIMIT_SCALE_v2
, user
);
526 pr_info_ratelimited("invalid rate from userspace: %llu\n",
532 static u64
user2rate_bytes(u32 user
)
536 r
= user
? U32_MAX
/ user
: U32_MAX
;
537 r
= (r
- 1) << XT_HASHLIMIT_BYTE_SHIFT
;
541 static void rateinfo_recalc(struct dsthash_ent
*dh
, unsigned long now
,
542 u32 mode
, int revision
)
544 unsigned long delta
= now
- dh
->rateinfo
.prev
;
550 if (revision
>= 3 && mode
& XT_HASHLIMIT_RATE_MATCH
) {
551 u64 interval
= dh
->rateinfo
.interval
* HZ
;
553 if (delta
< interval
)
556 dh
->rateinfo
.prev
= now
;
557 dh
->rateinfo
.prev_window
=
558 ((dh
->rateinfo
.current_rate
* interval
) >
559 (delta
* dh
->rateinfo
.rate
));
560 dh
->rateinfo
.current_rate
= 0;
565 dh
->rateinfo
.prev
= now
;
567 if (mode
& XT_HASHLIMIT_BYTES
) {
568 u64 tmp
= dh
->rateinfo
.credit
;
569 dh
->rateinfo
.credit
+= CREDITS_PER_JIFFY_BYTES
* delta
;
570 cap
= CREDITS_PER_JIFFY_BYTES
* HZ
;
571 if (tmp
>= dh
->rateinfo
.credit
) {/* overflow */
572 dh
->rateinfo
.credit
= cap
;
576 cpj
= (revision
== 1) ?
577 CREDITS_PER_JIFFY_v1
: CREDITS_PER_JIFFY
;
578 dh
->rateinfo
.credit
+= delta
* cpj
;
579 cap
= dh
->rateinfo
.credit_cap
;
581 if (dh
->rateinfo
.credit
> cap
)
582 dh
->rateinfo
.credit
= cap
;
585 static void rateinfo_init(struct dsthash_ent
*dh
,
586 struct xt_hashlimit_htable
*hinfo
, int revision
)
588 dh
->rateinfo
.prev
= jiffies
;
589 if (revision
>= 3 && hinfo
->cfg
.mode
& XT_HASHLIMIT_RATE_MATCH
) {
590 dh
->rateinfo
.prev_window
= 0;
591 dh
->rateinfo
.current_rate
= 0;
592 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_BYTES
) {
594 user2rate_bytes((u32
)hinfo
->cfg
.avg
);
595 if (hinfo
->cfg
.burst
)
597 hinfo
->cfg
.burst
* dh
->rateinfo
.rate
;
599 dh
->rateinfo
.burst
= dh
->rateinfo
.rate
;
601 dh
->rateinfo
.rate
= user2rate(hinfo
->cfg
.avg
);
603 hinfo
->cfg
.burst
+ dh
->rateinfo
.rate
;
605 dh
->rateinfo
.interval
= hinfo
->cfg
.interval
;
606 } else if (hinfo
->cfg
.mode
& XT_HASHLIMIT_BYTES
) {
607 dh
->rateinfo
.credit
= CREDITS_PER_JIFFY_BYTES
* HZ
;
608 dh
->rateinfo
.cost
= user2credits_byte(hinfo
->cfg
.avg
);
609 dh
->rateinfo
.credit_cap
= hinfo
->cfg
.burst
;
611 dh
->rateinfo
.credit
= user2credits(hinfo
->cfg
.avg
*
612 hinfo
->cfg
.burst
, revision
);
613 dh
->rateinfo
.cost
= user2credits(hinfo
->cfg
.avg
, revision
);
614 dh
->rateinfo
.credit_cap
= dh
->rateinfo
.credit
;
618 static inline __be32
maskl(__be32 a
, unsigned int l
)
620 return l
? htonl(ntohl(a
) & ~0 << (32 - l
)) : 0;
623 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
624 static void hashlimit_ipv6_mask(__be32
*i
, unsigned int p
)
628 i
[0] = maskl(i
[0], p
);
629 i
[1] = i
[2] = i
[3] = 0;
632 i
[1] = maskl(i
[1], p
- 32);
636 i
[2] = maskl(i
[2], p
- 64);
640 i
[3] = maskl(i
[3], p
- 96);
649 hashlimit_init_dst(const struct xt_hashlimit_htable
*hinfo
,
650 struct dsthash_dst
*dst
,
651 const struct sk_buff
*skb
, unsigned int protoff
)
653 __be16 _ports
[2], *ports
;
657 memset(dst
, 0, sizeof(*dst
));
659 switch (hinfo
->family
) {
661 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_HASH_DIP
)
662 dst
->ip
.dst
= maskl(ip_hdr(skb
)->daddr
,
664 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_HASH_SIP
)
665 dst
->ip
.src
= maskl(ip_hdr(skb
)->saddr
,
668 if (!(hinfo
->cfg
.mode
&
669 (XT_HASHLIMIT_HASH_DPT
| XT_HASHLIMIT_HASH_SPT
)))
671 nexthdr
= ip_hdr(skb
)->protocol
;
673 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
678 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_HASH_DIP
) {
679 memcpy(&dst
->ip6
.dst
, &ipv6_hdr(skb
)->daddr
,
680 sizeof(dst
->ip6
.dst
));
681 hashlimit_ipv6_mask(dst
->ip6
.dst
, hinfo
->cfg
.dstmask
);
683 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_HASH_SIP
) {
684 memcpy(&dst
->ip6
.src
, &ipv6_hdr(skb
)->saddr
,
685 sizeof(dst
->ip6
.src
));
686 hashlimit_ipv6_mask(dst
->ip6
.src
, hinfo
->cfg
.srcmask
);
689 if (!(hinfo
->cfg
.mode
&
690 (XT_HASHLIMIT_HASH_DPT
| XT_HASHLIMIT_HASH_SPT
)))
692 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
693 protoff
= ipv6_skip_exthdr(skb
, sizeof(struct ipv6hdr
), &nexthdr
, &frag_off
);
694 if ((int)protoff
< 0)
704 poff
= proto_ports_offset(nexthdr
);
706 ports
= skb_header_pointer(skb
, protoff
+ poff
, sizeof(_ports
),
709 _ports
[0] = _ports
[1] = 0;
714 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_HASH_SPT
)
715 dst
->src_port
= ports
[0];
716 if (hinfo
->cfg
.mode
& XT_HASHLIMIT_HASH_DPT
)
717 dst
->dst_port
= ports
[1];
721 static u32
hashlimit_byte_cost(unsigned int len
, struct dsthash_ent
*dh
)
723 u64 tmp
= xt_hashlimit_len_to_chunks(len
);
724 tmp
= tmp
* dh
->rateinfo
.cost
;
726 if (unlikely(tmp
> CREDITS_PER_JIFFY_BYTES
* HZ
))
727 tmp
= CREDITS_PER_JIFFY_BYTES
* HZ
;
729 if (dh
->rateinfo
.credit
< tmp
&& dh
->rateinfo
.credit_cap
) {
730 dh
->rateinfo
.credit_cap
--;
731 dh
->rateinfo
.credit
= CREDITS_PER_JIFFY_BYTES
* HZ
;
737 hashlimit_mt_common(const struct sk_buff
*skb
, struct xt_action_param
*par
,
738 struct xt_hashlimit_htable
*hinfo
,
739 const struct hashlimit_cfg3
*cfg
, int revision
)
741 unsigned long now
= jiffies
;
742 struct dsthash_ent
*dh
;
743 struct dsthash_dst dst
;
747 if (hashlimit_init_dst(hinfo
, &dst
, skb
, par
->thoff
) < 0)
751 dh
= dsthash_find(hinfo
, &dst
);
753 dh
= dsthash_alloc_init(hinfo
, &dst
, &race
);
758 /* Already got an entry, update expiration timeout */
759 dh
->expires
= now
+ msecs_to_jiffies(hinfo
->cfg
.expire
);
760 rateinfo_recalc(dh
, now
, hinfo
->cfg
.mode
, revision
);
762 dh
->expires
= jiffies
+ msecs_to_jiffies(hinfo
->cfg
.expire
);
763 rateinfo_init(dh
, hinfo
, revision
);
766 /* update expiration timeout */
767 dh
->expires
= now
+ msecs_to_jiffies(hinfo
->cfg
.expire
);
768 rateinfo_recalc(dh
, now
, hinfo
->cfg
.mode
, revision
);
771 if (cfg
->mode
& XT_HASHLIMIT_RATE_MATCH
) {
772 cost
= (cfg
->mode
& XT_HASHLIMIT_BYTES
) ? skb
->len
: 1;
773 dh
->rateinfo
.current_rate
+= cost
;
775 if (!dh
->rateinfo
.prev_window
&&
776 (dh
->rateinfo
.current_rate
<= dh
->rateinfo
.burst
)) {
777 spin_unlock(&dh
->lock
);
779 return !(cfg
->mode
& XT_HASHLIMIT_INVERT
);
785 if (cfg
->mode
& XT_HASHLIMIT_BYTES
)
786 cost
= hashlimit_byte_cost(skb
->len
, dh
);
788 cost
= dh
->rateinfo
.cost
;
790 if (dh
->rateinfo
.credit
>= cost
) {
791 /* below the limit */
792 dh
->rateinfo
.credit
-= cost
;
793 spin_unlock(&dh
->lock
);
795 return !(cfg
->mode
& XT_HASHLIMIT_INVERT
);
799 spin_unlock(&dh
->lock
);
801 /* default match is underlimit - so over the limit, we need to invert */
802 return cfg
->mode
& XT_HASHLIMIT_INVERT
;
810 hashlimit_mt_v1(const struct sk_buff
*skb
, struct xt_action_param
*par
)
812 const struct xt_hashlimit_mtinfo1
*info
= par
->matchinfo
;
813 struct xt_hashlimit_htable
*hinfo
= info
->hinfo
;
814 struct hashlimit_cfg3 cfg
= {};
817 ret
= cfg_copy(&cfg
, (void *)&info
->cfg
, 1);
822 return hashlimit_mt_common(skb
, par
, hinfo
, &cfg
, 1);
826 hashlimit_mt_v2(const struct sk_buff
*skb
, struct xt_action_param
*par
)
828 const struct xt_hashlimit_mtinfo2
*info
= par
->matchinfo
;
829 struct xt_hashlimit_htable
*hinfo
= info
->hinfo
;
830 struct hashlimit_cfg3 cfg
= {};
833 ret
= cfg_copy(&cfg
, (void *)&info
->cfg
, 2);
838 return hashlimit_mt_common(skb
, par
, hinfo
, &cfg
, 2);
842 hashlimit_mt(const struct sk_buff
*skb
, struct xt_action_param
*par
)
844 const struct xt_hashlimit_mtinfo3
*info
= par
->matchinfo
;
845 struct xt_hashlimit_htable
*hinfo
= info
->hinfo
;
847 return hashlimit_mt_common(skb
, par
, hinfo
, &info
->cfg
, 3);
850 static int hashlimit_mt_check_common(const struct xt_mtchk_param
*par
,
851 struct xt_hashlimit_htable
**hinfo
,
852 struct hashlimit_cfg3
*cfg
,
853 const char *name
, int revision
)
855 struct net
*net
= par
->net
;
858 if (cfg
->gc_interval
== 0 || cfg
->expire
== 0)
860 if (par
->family
== NFPROTO_IPV4
) {
861 if (cfg
->srcmask
> 32 || cfg
->dstmask
> 32)
864 if (cfg
->srcmask
> 128 || cfg
->dstmask
> 128)
868 if (cfg
->mode
& ~XT_HASHLIMIT_ALL
) {
869 pr_info_ratelimited("Unknown mode mask %X, kernel too old?\n",
874 /* Check for overflow. */
875 if (revision
>= 3 && cfg
->mode
& XT_HASHLIMIT_RATE_MATCH
) {
876 if (cfg
->avg
== 0 || cfg
->avg
> U32_MAX
) {
877 pr_info_ratelimited("invalid rate\n");
881 if (cfg
->interval
== 0) {
882 pr_info_ratelimited("invalid interval\n");
885 } else if (cfg
->mode
& XT_HASHLIMIT_BYTES
) {
886 if (user2credits_byte(cfg
->avg
) == 0) {
887 pr_info_ratelimited("overflow, rate too high: %llu\n",
891 } else if (cfg
->burst
== 0 ||
892 user2credits(cfg
->avg
* cfg
->burst
, revision
) <
893 user2credits(cfg
->avg
, revision
)) {
894 pr_info_ratelimited("overflow, try lower: %llu/%llu\n",
895 cfg
->avg
, cfg
->burst
);
899 mutex_lock(&hashlimit_mutex
);
900 *hinfo
= htable_find_get(net
, name
, par
->family
);
901 if (*hinfo
== NULL
) {
902 ret
= htable_create(net
, cfg
, name
, par
->family
,
905 mutex_unlock(&hashlimit_mutex
);
909 mutex_unlock(&hashlimit_mutex
);
914 static int hashlimit_mt_check_v1(const struct xt_mtchk_param
*par
)
916 struct xt_hashlimit_mtinfo1
*info
= par
->matchinfo
;
917 struct hashlimit_cfg3 cfg
= {};
920 if (info
->name
[sizeof(info
->name
) - 1] != '\0')
923 ret
= cfg_copy(&cfg
, (void *)&info
->cfg
, 1);
928 return hashlimit_mt_check_common(par
, &info
->hinfo
,
929 &cfg
, info
->name
, 1);
932 static int hashlimit_mt_check_v2(const struct xt_mtchk_param
*par
)
934 struct xt_hashlimit_mtinfo2
*info
= par
->matchinfo
;
935 struct hashlimit_cfg3 cfg
= {};
938 if (info
->name
[sizeof(info
->name
) - 1] != '\0')
941 ret
= cfg_copy(&cfg
, (void *)&info
->cfg
, 2);
946 return hashlimit_mt_check_common(par
, &info
->hinfo
,
947 &cfg
, info
->name
, 2);
950 static int hashlimit_mt_check(const struct xt_mtchk_param
*par
)
952 struct xt_hashlimit_mtinfo3
*info
= par
->matchinfo
;
954 if (info
->name
[sizeof(info
->name
) - 1] != '\0')
957 return hashlimit_mt_check_common(par
, &info
->hinfo
, &info
->cfg
,
961 static void hashlimit_mt_destroy_v2(const struct xt_mtdtor_param
*par
)
963 const struct xt_hashlimit_mtinfo2
*info
= par
->matchinfo
;
965 htable_put(info
->hinfo
);
968 static void hashlimit_mt_destroy_v1(const struct xt_mtdtor_param
*par
)
970 const struct xt_hashlimit_mtinfo1
*info
= par
->matchinfo
;
972 htable_put(info
->hinfo
);
975 static void hashlimit_mt_destroy(const struct xt_mtdtor_param
*par
)
977 const struct xt_hashlimit_mtinfo3
*info
= par
->matchinfo
;
979 htable_put(info
->hinfo
);
982 static struct xt_match hashlimit_mt_reg
[] __read_mostly
= {
986 .family
= NFPROTO_IPV4
,
987 .match
= hashlimit_mt_v1
,
988 .matchsize
= sizeof(struct xt_hashlimit_mtinfo1
),
989 .usersize
= offsetof(struct xt_hashlimit_mtinfo1
, hinfo
),
990 .checkentry
= hashlimit_mt_check_v1
,
991 .destroy
= hashlimit_mt_destroy_v1
,
997 .family
= NFPROTO_IPV4
,
998 .match
= hashlimit_mt_v2
,
999 .matchsize
= sizeof(struct xt_hashlimit_mtinfo2
),
1000 .usersize
= offsetof(struct xt_hashlimit_mtinfo2
, hinfo
),
1001 .checkentry
= hashlimit_mt_check_v2
,
1002 .destroy
= hashlimit_mt_destroy_v2
,
1006 .name
= "hashlimit",
1008 .family
= NFPROTO_IPV4
,
1009 .match
= hashlimit_mt
,
1010 .matchsize
= sizeof(struct xt_hashlimit_mtinfo3
),
1011 .usersize
= offsetof(struct xt_hashlimit_mtinfo3
, hinfo
),
1012 .checkentry
= hashlimit_mt_check
,
1013 .destroy
= hashlimit_mt_destroy
,
1016 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1018 .name
= "hashlimit",
1020 .family
= NFPROTO_IPV6
,
1021 .match
= hashlimit_mt_v1
,
1022 .matchsize
= sizeof(struct xt_hashlimit_mtinfo1
),
1023 .usersize
= offsetof(struct xt_hashlimit_mtinfo1
, hinfo
),
1024 .checkentry
= hashlimit_mt_check_v1
,
1025 .destroy
= hashlimit_mt_destroy_v1
,
1029 .name
= "hashlimit",
1031 .family
= NFPROTO_IPV6
,
1032 .match
= hashlimit_mt_v2
,
1033 .matchsize
= sizeof(struct xt_hashlimit_mtinfo2
),
1034 .usersize
= offsetof(struct xt_hashlimit_mtinfo2
, hinfo
),
1035 .checkentry
= hashlimit_mt_check_v2
,
1036 .destroy
= hashlimit_mt_destroy_v2
,
1040 .name
= "hashlimit",
1042 .family
= NFPROTO_IPV6
,
1043 .match
= hashlimit_mt
,
1044 .matchsize
= sizeof(struct xt_hashlimit_mtinfo3
),
1045 .usersize
= offsetof(struct xt_hashlimit_mtinfo3
, hinfo
),
1046 .checkentry
= hashlimit_mt_check
,
1047 .destroy
= hashlimit_mt_destroy
,
1054 static void *dl_seq_start(struct seq_file
*s
, loff_t
*pos
)
1055 __acquires(htable
->lock
)
1057 struct xt_hashlimit_htable
*htable
= s
->private;
1058 unsigned int *bucket
;
1060 spin_lock_bh(&htable
->lock
);
1061 if (*pos
>= htable
->cfg
.size
)
1064 bucket
= kmalloc(sizeof(unsigned int), GFP_ATOMIC
);
1066 return ERR_PTR(-ENOMEM
);
1072 static void *dl_seq_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
1074 struct xt_hashlimit_htable
*htable
= s
->private;
1075 unsigned int *bucket
= v
;
1078 if (*pos
>= htable
->cfg
.size
) {
1085 static void dl_seq_stop(struct seq_file
*s
, void *v
)
1086 __releases(htable
->lock
)
1088 struct xt_hashlimit_htable
*htable
= s
->private;
1089 unsigned int *bucket
= v
;
1091 if (!IS_ERR(bucket
))
1093 spin_unlock_bh(&htable
->lock
);
1096 static void dl_seq_print(struct dsthash_ent
*ent
, u_int8_t family
,
1101 seq_printf(s
, "%ld %pI4:%u->%pI4:%u %llu %llu %llu\n",
1102 (long)(ent
->expires
- jiffies
)/HZ
,
1104 ntohs(ent
->dst
.src_port
),
1106 ntohs(ent
->dst
.dst_port
),
1107 ent
->rateinfo
.credit
, ent
->rateinfo
.credit_cap
,
1108 ent
->rateinfo
.cost
);
1110 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1112 seq_printf(s
, "%ld %pI6:%u->%pI6:%u %llu %llu %llu\n",
1113 (long)(ent
->expires
- jiffies
)/HZ
,
1115 ntohs(ent
->dst
.src_port
),
1117 ntohs(ent
->dst
.dst_port
),
1118 ent
->rateinfo
.credit
, ent
->rateinfo
.credit_cap
,
1119 ent
->rateinfo
.cost
);
1127 static int dl_seq_real_show_v2(struct dsthash_ent
*ent
, u_int8_t family
,
1130 const struct xt_hashlimit_htable
*ht
= s
->private;
1132 spin_lock(&ent
->lock
);
1133 /* recalculate to show accurate numbers */
1134 rateinfo_recalc(ent
, jiffies
, ht
->cfg
.mode
, 2);
1136 dl_seq_print(ent
, family
, s
);
1138 spin_unlock(&ent
->lock
);
1139 return seq_has_overflowed(s
);
1142 static int dl_seq_real_show_v1(struct dsthash_ent
*ent
, u_int8_t family
,
1145 const struct xt_hashlimit_htable
*ht
= s
->private;
1147 spin_lock(&ent
->lock
);
1148 /* recalculate to show accurate numbers */
1149 rateinfo_recalc(ent
, jiffies
, ht
->cfg
.mode
, 1);
1151 dl_seq_print(ent
, family
, s
);
1153 spin_unlock(&ent
->lock
);
1154 return seq_has_overflowed(s
);
1157 static int dl_seq_real_show(struct dsthash_ent
*ent
, u_int8_t family
,
1160 const struct xt_hashlimit_htable
*ht
= s
->private;
1162 spin_lock(&ent
->lock
);
1163 /* recalculate to show accurate numbers */
1164 rateinfo_recalc(ent
, jiffies
, ht
->cfg
.mode
, 3);
1166 dl_seq_print(ent
, family
, s
);
1168 spin_unlock(&ent
->lock
);
1169 return seq_has_overflowed(s
);
1172 static int dl_seq_show_v2(struct seq_file
*s
, void *v
)
1174 struct xt_hashlimit_htable
*htable
= s
->private;
1175 unsigned int *bucket
= (unsigned int *)v
;
1176 struct dsthash_ent
*ent
;
1178 if (!hlist_empty(&htable
->hash
[*bucket
])) {
1179 hlist_for_each_entry(ent
, &htable
->hash
[*bucket
], node
)
1180 if (dl_seq_real_show_v2(ent
, htable
->family
, s
))
1186 static int dl_seq_show_v1(struct seq_file
*s
, void *v
)
1188 struct xt_hashlimit_htable
*htable
= s
->private;
1189 unsigned int *bucket
= v
;
1190 struct dsthash_ent
*ent
;
1192 if (!hlist_empty(&htable
->hash
[*bucket
])) {
1193 hlist_for_each_entry(ent
, &htable
->hash
[*bucket
], node
)
1194 if (dl_seq_real_show_v1(ent
, htable
->family
, s
))
1200 static int dl_seq_show(struct seq_file
*s
, void *v
)
1202 struct xt_hashlimit_htable
*htable
= s
->private;
1203 unsigned int *bucket
= v
;
1204 struct dsthash_ent
*ent
;
1206 if (!hlist_empty(&htable
->hash
[*bucket
])) {
1207 hlist_for_each_entry(ent
, &htable
->hash
[*bucket
], node
)
1208 if (dl_seq_real_show(ent
, htable
->family
, s
))
1214 static const struct seq_operations dl_seq_ops_v1
= {
1215 .start
= dl_seq_start
,
1216 .next
= dl_seq_next
,
1217 .stop
= dl_seq_stop
,
1218 .show
= dl_seq_show_v1
1221 static const struct seq_operations dl_seq_ops_v2
= {
1222 .start
= dl_seq_start
,
1223 .next
= dl_seq_next
,
1224 .stop
= dl_seq_stop
,
1225 .show
= dl_seq_show_v2
1228 static const struct seq_operations dl_seq_ops
= {
1229 .start
= dl_seq_start
,
1230 .next
= dl_seq_next
,
1231 .stop
= dl_seq_stop
,
1235 static int dl_proc_open_v2(struct inode
*inode
, struct file
*file
)
1237 int ret
= seq_open(file
, &dl_seq_ops_v2
);
1240 struct seq_file
*sf
= file
->private_data
;
1242 sf
->private = PDE_DATA(inode
);
1247 static int dl_proc_open_v1(struct inode
*inode
, struct file
*file
)
1249 int ret
= seq_open(file
, &dl_seq_ops_v1
);
1252 struct seq_file
*sf
= file
->private_data
;
1253 sf
->private = PDE_DATA(inode
);
1258 static int dl_proc_open(struct inode
*inode
, struct file
*file
)
1260 int ret
= seq_open(file
, &dl_seq_ops
);
1263 struct seq_file
*sf
= file
->private_data
;
1265 sf
->private = PDE_DATA(inode
);
1270 static const struct file_operations dl_file_ops_v2
= {
1271 .open
= dl_proc_open_v2
,
1273 .llseek
= seq_lseek
,
1274 .release
= seq_release
1277 static const struct file_operations dl_file_ops_v1
= {
1278 .open
= dl_proc_open_v1
,
1280 .llseek
= seq_lseek
,
1281 .release
= seq_release
1284 static const struct file_operations dl_file_ops
= {
1285 .open
= dl_proc_open
,
1287 .llseek
= seq_lseek
,
1288 .release
= seq_release
1291 static int __net_init
hashlimit_proc_net_init(struct net
*net
)
1293 struct hashlimit_net
*hashlimit_net
= hashlimit_pernet(net
);
1295 hashlimit_net
->ipt_hashlimit
= proc_mkdir("ipt_hashlimit", net
->proc_net
);
1296 if (!hashlimit_net
->ipt_hashlimit
)
1298 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1299 hashlimit_net
->ip6t_hashlimit
= proc_mkdir("ip6t_hashlimit", net
->proc_net
);
1300 if (!hashlimit_net
->ip6t_hashlimit
) {
1301 remove_proc_entry("ipt_hashlimit", net
->proc_net
);
1308 static void __net_exit
hashlimit_proc_net_exit(struct net
*net
)
1310 struct xt_hashlimit_htable
*hinfo
;
1311 struct hashlimit_net
*hashlimit_net
= hashlimit_pernet(net
);
1313 /* hashlimit_net_exit() is called before hashlimit_mt_destroy().
1314 * Make sure that the parent ipt_hashlimit and ip6t_hashlimit proc
1315 * entries is empty before trying to remove it.
1317 mutex_lock(&hashlimit_mutex
);
1318 hlist_for_each_entry(hinfo
, &hashlimit_net
->htables
, node
)
1319 htable_remove_proc_entry(hinfo
);
1320 hashlimit_net
->ipt_hashlimit
= NULL
;
1321 hashlimit_net
->ip6t_hashlimit
= NULL
;
1322 mutex_unlock(&hashlimit_mutex
);
1324 remove_proc_entry("ipt_hashlimit", net
->proc_net
);
1325 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
1326 remove_proc_entry("ip6t_hashlimit", net
->proc_net
);
1330 static int __net_init
hashlimit_net_init(struct net
*net
)
1332 struct hashlimit_net
*hashlimit_net
= hashlimit_pernet(net
);
1334 INIT_HLIST_HEAD(&hashlimit_net
->htables
);
1335 return hashlimit_proc_net_init(net
);
1338 static void __net_exit
hashlimit_net_exit(struct net
*net
)
1340 hashlimit_proc_net_exit(net
);
1343 static struct pernet_operations hashlimit_net_ops
= {
1344 .init
= hashlimit_net_init
,
1345 .exit
= hashlimit_net_exit
,
1346 .id
= &hashlimit_net_id
,
1347 .size
= sizeof(struct hashlimit_net
),
1350 static int __init
hashlimit_mt_init(void)
1354 err
= register_pernet_subsys(&hashlimit_net_ops
);
1357 err
= xt_register_matches(hashlimit_mt_reg
,
1358 ARRAY_SIZE(hashlimit_mt_reg
));
1363 hashlimit_cachep
= kmem_cache_create("xt_hashlimit",
1364 sizeof(struct dsthash_ent
), 0, 0,
1366 if (!hashlimit_cachep
) {
1367 pr_warn("unable to create slab cache\n");
1373 xt_unregister_matches(hashlimit_mt_reg
, ARRAY_SIZE(hashlimit_mt_reg
));
1375 unregister_pernet_subsys(&hashlimit_net_ops
);
1380 static void __exit
hashlimit_mt_exit(void)
1382 xt_unregister_matches(hashlimit_mt_reg
, ARRAY_SIZE(hashlimit_mt_reg
));
1383 unregister_pernet_subsys(&hashlimit_net_ops
);
1386 kmem_cache_destroy(hashlimit_cachep
);
1389 module_init(hashlimit_mt_init
);
1390 module_exit(hashlimit_mt_exit
);